local_motivation.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "682e768d",
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import glob\n",
    "import os\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import json\n",
    "import seaborn as sns\n",
    "import pickle as pkl\n",
    "import copy\n",
    "import matplotlib.patches as mpatches\n",
    "from utils import OPT_PATH as opt_path\n",
    "# refer to https://matplotlib.org/stable/api/markers_api.html\n",
    "markers = [\"o\", \"v\", \"s\", \"*\", \"D\"]\n",
    "# refer to https://matplotlib.org/stable/gallery/color/named_colors.html\n",
    "colors = ['tab:blue', 'tab:orange', 'tab:green', 'tab:red', 'tab:purple', 'tab:brown', 'tab:pink', 'tab:gray']\n",
    "# https://matplotlib.org/3.1.0/gallery/lines_bars_and_markers/linestyles.html\n",
    "linestyles = 'solid'\n",
    "# refer to https://github.com/bvalgard/Boxplots/blob/master/Boxplots.ipynb"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3e4de0ef",
   "metadata": {},
   "source": [
    "## Universality v.s. Transferability"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4e0b03a5",
   "metadata": {},
   "source": [
    "### I-FGSM"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "75172213",
   "metadata": {},
   "outputs": [],
   "source": [
    "path = glob.glob(os.path.join(opt_path, '*-DTMI-*motivation_IFGSM/eval_universal.npy'))\n",
    "json_path = glob.glob(os.path.join(opt_path, '*-DTMI-*motivation_IFGSM/pert_to_OriImageInd.json'))\n",
    "eval_universal = np.load(path[0])\n",
    "with open(json_path[0], 'r') as ipt:\n",
    "    pert_to_OriImageInd = json.load(ipt)\n",
    "# obtain each pert's white-box universal \n",
    "ifgsm_white_universal = []\n",
    "for pert_ind in range(1000):\n",
    "    preds = eval_universal[2][pert_ind]\n",
    "    image_id = pert_to_OriImageInd[str(pert_ind)]\n",
    "    preds = np.delete(preds, image_id)\n",
    "    ifgsm_white_universal.append(np.sum(preds))\n",
    "ifgsm_white_universal = np.array(ifgsm_white_universal)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a3250e54",
   "metadata": {},
   "source": [
    "### DTMI"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "id": "c9c162e5",
   "metadata": {},
   "outputs": [],
   "source": [
    "path = glob.glob(os.path.join(opt_path, '*-DTMI-*motivation/eval_universal.npy'))\n",
    "json_path = glob.glob(os.path.join(opt_path, '*-DTMI-*motivation/pert_to_OriImageInd.json'))\n",
    "eval_universal = np.load(path[0])\n",
    "with open(json_path[0], 'r') as ipt:\n",
    "    pert_to_OriImageInd = json.load(ipt)\n",
    "# obtain each pert's white-box universal \n",
    "dtmi_white_universal = []\n",
    "for pert_ind in range(1000):\n",
    "    preds = eval_universal[2][pert_ind]\n",
    "    image_id = pert_to_OriImageInd[str(pert_ind)]\n",
    "    preds = np.delete(preds, image_id)\n",
    "    dtmi_white_universal.append(np.sum(preds))\n",
    "dtmi_white_universal = np.array(dtmi_white_universal)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "id": "9478b2ae",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 288x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig = plt.figure(figsize=(4,3)) \n",
    "ax1 = fig.add_subplot(111)\n",
    "ind = np.argsort(-ifgsm_white_universal)\n",
    "ax1.bar(range(len(ind)), np.array(ifgsm_white_universal)[ind], width=1, alpha=1, color=sns.color_palette(\"Paired\")[5])\n",
    "# ax1.set_yticks([0.00, 0.10, 0.20, 0.30, 0.40, 0.50])\n",
    "# ax2 = fig.add_subplot(312)\n",
    "ax1.bar(range(len(ind)), np.array(dtmi_white_universal)[ind], width=1, alpha=0.4, color=sns.color_palette(\"Paired\")[1])\n",
    "# ax3 = fig.add_subplot(313)\n",
    "# ax3.bar(range(len(ind)), np.array(local_dtmi_white_universal)[ind], width=1, alpha=0.5, color='orange')\n",
    "\n",
    "\n",
    "labels = ['I-FGSM', 'DTMI'] \n",
    "color = [sns.color_palette(\"Paired\")[5], sns.color_palette(\"Paired\")[1]]\n",
    "patches = [ mpatches.Patch(color=color[i], label=\"{:s}\".format(labels[i]) ) for i in range(len(color)) ] \n",
    "ax=plt.gca()\n",
    "box = ax.get_position()\n",
    "# ax.set_position([box.x0, box.y0, box.width , box.height* 0.8])\n",
    "ax.legend(handles=patches, fontsize=10) #生成legend\n",
    "plt.xlabel('Generated adversarial perturbations')\n",
    "plt.ylabel('Successful attacks on other images')\n",
    "# plt.show()\n",
    "plt.savefig('./figures/perturbation_universality.pdf', format='pdf', bbox_inches = 'tight')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2f6428d1",
   "metadata": {},
   "source": [
    "### SI "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "9bdb1f42",
   "metadata": {},
   "outputs": [],
   "source": [
    "path = glob.glob(os.path.join('/share_io02_ssd/zhipeng/targeted_attack_loss/', 'Loss-resnet50-NIPSDataset-SI_DTMI-Target_CE-baseline/eval_universal.npy'))\n",
    "json_path = glob.glob(os.path.join('/share_io02_ssd/zhipeng/targeted_attack_loss/', 'Loss-resnet50-NIPSDataset-SI_DTMI-Target_CE-baseline/pert_to_OriImageInd.json'))\n",
    "eval_universal = np.load(path[0])\n",
    "with open(json_path[0], 'r') as ipt:\n",
    "    pert_to_OriImageInd = json.load(ipt)\n",
    "# obtain each pert's white-box universal \n",
    "SI_white_universal = []\n",
    "for pert_ind in range(1000):\n",
    "    preds = eval_universal[2][pert_ind]\n",
    "    image_id = pert_to_OriImageInd[str(pert_ind)]\n",
    "    preds = np.delete(preds, image_id)\n",
    "    SI_white_universal.append(np.sum(preds))\n",
    "SI_white_universal = np.array(SI_white_universal)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "c34f2c01",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0, 0.5, 'Successful attacks on other images')"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 288x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig = plt.figure(figsize=(4,3)) \n",
    "ax1 = fig.add_subplot(111)\n",
    "ind = np.argsort(-ifgsm_white_universal)\n",
    "ax1.bar(range(len(ind)), np.array(ifgsm_white_universal)[ind], width=1, alpha=1, color=sns.color_palette(\"Paired\")[5])\n",
    "# ax1.set_yticks([0.00, 0.10, 0.20, 0.30, 0.40, 0.50])\n",
    "# ax2 = fig.add_subplot(312)\n",
    "ax1.bar(range(len(ind)), np.array(SI_white_universal)[ind], width=1, alpha=0.4, color=sns.color_palette(\"Paired\")[1])\n",
    "# ax3 = fig.add_subplot(313)\n",
    "# ax3.bar(range(len(ind)), np.array(local_dtmi_white_universal)[ind], width=1, alpha=0.5, color='orange')\n",
    "\n",
    "\n",
    "labels = ['I-FGSM', 'DTMI'] \n",
    "color = [sns.color_palette(\"Paired\")[5], sns.color_palette(\"Paired\")[1]]\n",
    "patches = [ mpatches.Patch(color=color[i], label=\"{:s}\".format(labels[i]) ) for i in range(len(color)) ] \n",
    "ax=plt.gca()\n",
    "box = ax.get_position()\n",
    "# ax.set_position([box.x0, box.y0, box.width , box.height* 0.8])\n",
    "ax.legend(handles=patches, fontsize=10) #生成legend\n",
    "plt.xlabel('Generated adversarial perturbations')\n",
    "plt.ylabel('Successful attacks on other images')\n",
    "# plt.show()\n",
    "# plt.savefig('./figures/perturbation_universality.pdf', format='pdf', bbox_inches = 'tight')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e14463ba",
   "metadata": {},
   "source": [
    "### Admix"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "9f980f62",
   "metadata": {},
   "outputs": [],
   "source": [
    "path = glob.glob(os.path.join('/share_io02_ssd/zhipeng/targeted_attack_loss/', 'Loss-resnet50-NIPSDataset-Admix_DTMI-Target_CE-baseline/eval_universal.npy'))\n",
    "json_path = glob.glob(os.path.join('/share_io02_ssd/zhipeng/targeted_attack_loss/', 'Loss-resnet50-NIPSDataset-Admix_DTMI-Target_CE-baseline/pert_to_OriImageInd.json'))\n",
    "eval_universal = np.load(path[0])\n",
    "with open(json_path[0], 'r') as ipt:\n",
    "    pert_to_OriImageInd = json.load(ipt)\n",
    "# obtain each pert's white-box universal \n",
    "admix_white_universal = []\n",
    "for pert_ind in range(1000):\n",
    "    preds = eval_universal[2][pert_ind]\n",
    "    image_id = pert_to_OriImageInd[str(pert_ind)]\n",
    "    preds = np.delete(preds, image_id)\n",
    "    admix_white_universal.append(np.sum(preds))\n",
    "admix_white_universal = np.array(admix_white_universal)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "74268414",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0, 0.5, 'Successful attacks on other images')"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 288x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig = plt.figure(figsize=(4,3)) \n",
    "ax1 = fig.add_subplot(111)\n",
    "ind = np.argsort(-ifgsm_white_universal)\n",
    "ax1.bar(range(len(ind)), np.array(ifgsm_white_universal)[ind], width=1, alpha=1, color=sns.color_palette(\"Paired\")[5])\n",
    "# ax1.set_yticks([0.00, 0.10, 0.20, 0.30, 0.40, 0.50])\n",
    "# ax2 = fig.add_subplot(312)\n",
    "ax1.bar(range(len(ind)), np.array(admix_white_universal)[ind], width=1, alpha=0.4, color=sns.color_palette(\"Paired\")[1])\n",
    "# ax3 = fig.add_subplot(313)\n",
    "# ax3.bar(range(len(ind)), np.array(local_dtmi_white_universal)[ind], width=1, alpha=0.5, color='orange')\n",
    "\n",
    "\n",
    "labels = ['I-FGSM', 'DTMI'] \n",
    "color = [sns.color_palette(\"Paired\")[5], sns.color_palette(\"Paired\")[1]]\n",
    "patches = [ mpatches.Patch(color=color[i], label=\"{:s}\".format(labels[i]) ) for i in range(len(color)) ] \n",
    "ax=plt.gca()\n",
    "box = ax.get_position()\n",
    "# ax.set_position([box.x0, box.y0, box.width , box.height* 0.8])\n",
    "ax.legend(handles=patches, fontsize=10) #生成legend\n",
    "plt.xlabel('Generated adversarial perturbations')\n",
    "plt.ylabel('Successful attacks on other images')\n",
    "# plt.show()\n",
    "# plt.savefig('./figures/perturbation_universality.pdf', format='pdf', bbox_inches = 'tight')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "800e0f1e",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.11"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}